The Regional Distribution of Skill Premia in Urban China
and Implications for Growth and Inequality
John Whalley
University of Western Ontario
Center for International Governance Innovation (CIGI)
NBER
Chunbing Xing
Beijing Normal University
March, 2013

We are grateful to the Ontario Research Fund for support, and to Shi Li, Terry Sicular, Lance Lochner, Todd
Stinebrickner, Chunding Li, Xiaojun Shi, Xiliang Zhao, Yunfei Wang, and seminar participants at the University of
Western Ontario, SEBA-GATE 3rd Annual Workshop, Shenyang Agricultural University, the 2012 RUSE China
Workshop at Jinan University, and the referee for helpful comments. Chunbing Xing is also grateful to the
National Natural Science Foundation of China for financial support (No. 71103019).
1
The Regional Distribution of Skill Premia in Urban China
and Implications for Growth and Inequality
Abstract. We document and discuss the implications of a sharp increase in the
regional dispersion of skill premia in China in recent years. This has previously been
little noted or discussed. We use three urban household surveys for 1995, 2002, and
2007 and estimate skill premia at provincial and city levels. Results show an increase
in the skill premium across all regions between 1995 and 2002, but only coastal
regions show significant increases in skill premia between 2002 and 2007. For 2007,
coastal regions also have much higher within-region wage inequality, and this
contributes more to overall urban wage inequality than within-region inequality of
non-coastal regions. Using a fixed effects model at city level, we find that ownership
restructuring is a significant factor in driving up skill premia during the first period
(1995 to 2002), and that the ongoing process of China’s integration into the global
economy plays a significant and regionally concentrated role in the second period
(2002 to 2007). Finally, we suggest that the Hukou registration system and rising
housing prices may prevent skilled labour from moving to high skill premium regions
and retard mobility-induced reductions in inequality. This effect may also retard even
higher growth.
Key Words: Skill Premia, Regional Distribution, Inequality, Urban China
JEL Classification: I24; J24; R23
1
1. Introduction
The behaviour of the skill premium (the wage differential between skilled and
unskilled labour) is central to an understanding of the evolution of inequality in urban
China. In this paper we use education as a measure of skill, and use the terms skill
premium and education premium (or the return to education) interchangeably.1 There
has been a large amount of research aiming to estimate the skill premium over time
(see Zhang et al., 2005; Li and Ding, 2003).2 How the skill premia have evolved
within regions and how this may have affected overall inequality is, however, largely
neglected in literature.
One major reason for our investigating the regional dimension of skill premia is
China's large regional gap in economic development. Between 1995 and 2007, coastal
provinces had higher rates of growth than hinterland provinces. In 2007, the GDP per
capita of the richest province, Shanghai, was nearly 10 times that of the poorest,
Guizhou (National Bureau of Statistics, 2008). Regional income inequality has thus
become a major contributor to China's overall inequality, and regional imbalance has
also become a threat to the sustainability of China's high levels of growth. While the
regional income gap has been studied extensively, little attention has been given to
differences in skill premia across regions, which is another important dimension of an
1
In this paper we use education levels to define a dummy variable indicating whether a worker is skilled or
unskilled. Since data on years of schooling and education levels is available, we also use this data directly.
2
There are also many estimates of the return to education across other divides. Dong and Bowles (2002), for
instance, estimate the return to education by ownership.
2
understanding of overall inequality, which also has implications for the mobility of
workers with different education levels.
Here, we use three Chinese urban household surveys, from 1995, 2002, and 2007,
to estimate skill premia at a provincial and prefecture (or city) level for each year.
Provinces are the highest-level administrative divisions in China, and prefecture-level
cities (hereafter cities) are administrative subdivisions of provincial-level divisions.
Municipalities directly under the central government (Beijing, Chongqing, and
Shanghai)3 are treated as both provinces and cities in the estimation. We often use the
term "region" to encompass these various interpretations. We find substantial yearly
dispersion across regions, but also find an increase in skill premia across all regions
between 1995 and 2002, while only coastal regions show significant increases in skill
premia between 2002 and 2007.
We construct a panel dataset at city level and use a fixed effects model to
determine which factors account for the increase in skill premia. In the 1995 to 2002
period, institutional change (i.e., ownership restructuring) seemingly plays a key role
in driving up the skill premia in the various regions. We find a significant relationship
between export activity and increases in skill premia in the 2002 to 2007 period,
suggesting that China’s entry into the World Trade Organisation (WTO) and its wider
integration into the global economy after 2001 may have played a significant role. We
also discuss the possibility that high migration rates of unskilled labour slow the
growth of real wages and drive up skill premia in coastal regions. We only find weak
3
Tianjin, another municipality, was not covered by the survey.
3
evidence for this hypothesis.
We suggest that investigating how the regional distribution of skill premia
changes will provide a deeper understanding of changes in urban wage inequality. If
wage inequality is decomposed into between-region and within-region inequality, the
within-region inequality can, in turn, be expressed by a weighted average of the
inequality for all regions. The various regions in China show different inequalities and,
hence, different contributions to overall urban wage inequality. We find that, by 2007,
coastal regions have much higher within-region inequality (because of higher skill
premia), and this contributes more to overall urban wage inequality than non-coastal
regions, a feature that has not yet been documented in the literature.
An investigation of the regional distribution of skill premia also has implications
for growth. In recent years a growing literature emphasises resource misallocation in
explaining low economic efficiency (Hsieh and Klenow, 2009).4 In this paper, we
view the dispersion of skill premia as a measure of labour misallocation across
regions. A wider premia dispersion indicates a greater difference in relative marginal
productivity across regions; therefore, a reallocation of labour (through migration)
could improve efficiency and promote economic growth.
A more complete picture of urban inequality and the relationship between
economic transition and skill premia also has policy implications. As coastal regions
have much higher skill premia, which contributes to overall urban wage inequality,
4
Earlier research focusing on regional inequality and growth considers income differentials across regions and the
convergence (or divergence) in income levels in the growth process (Chen and Belton, 1996).
4
the mobility of educated workers, relaxation of the household registration system (i.e.,
the Hukou system, which restricts population mobility),5 and an increased supply of
urban housing in coastal regions could help reduce overall inequality and increase the
growth potential of China’s economy. These policies should be prioritised given the
sharp increase in the number of college graduates following the initiation of China's
higher education expansion in the late 1990s. Local education or training programmes
and employer-based training may help increase supply of skilled workers and reduce
skill differentials in high skill premia regions. Finally, investing in and offering
preferential policies to low premium regions may also help equalise regional skill
premia, but these often involve complicated political processes such as negotiations
between central and local governments and competition among local governments.
Indeed, whether the skill premia dispersion and inequality can be reduced will depend
on local circumstances, technology, and the nature of the policies.
This paper is organised as follows: We discuss background and literature on skill
premia in section 2; we describe our data in section 3; we report estimates of skill
premia, their regional distribution, and evolution over time in section 4; we
investigate the factors underlying patterns of regional skill premia over time and
discuss the implications for overall urban wage inequality and potential policy actions
in section 5; our conclusions are given in section 6.
2. Why is the regional distribution of skill premia important in China?
5
A more detailed discussion of the Hukou system is given in the next section.
5
The Chinese economy grew rapidly between 1995 and 2007 at an average annual
growth rate of 10 percent. The 2002 to 2007 period experienced relatively higher
growth (National Bureau of Statistics, 2009: Table 2-4). Significant changes occurred
in the economy during the 1995 to 2007 period. In the late 1990s, many small- to
medium-sized state-owned enterprises (SOEs) in the urban sector were privatised
(Appleton et al., 2002; Solinger, 2002), and more workers were allocated to the
private sector. Following China's entry into the WTO in 2001, Chinese exports
increased sharply and became a major contributor to the country's economic growth.
Other major changes in the labour market included the expansion of higher education,
starting from 1999 (Li et al., 2011), and large increases in rural-to-urban migration
(Cai et al., 2009). Increasing inequality accompanied China's rapid economic growth
(World Bank, 2009).
As a large developing country in economic transition, China's various regions
have very different levels of economic development. Earlier research has concluded
that regional disparity is a major contributor to overall inequality (Gustafsson et al.
2008). The literature has, however, failed to consider that the relative wages of skilled
and unskilled labour may be different across regions. Given the considerable effort
devoted to estimating skill premia, the extension to include the regional dimension is
a natural next step.
Institutional factors may play a role in causing skill premia differentials among
regions. For example, wage determination mechanisms were significantly different
across firm ownership types in the 1990s, with private sector firms sharing a higher
6
skill premium (Meng, 2000; Dong and Bowles, 2002; Xing, 2008; Zhang et al., 2005).
Different regions also had a different firm ownership mix. In 2008, for example, the
registered employment share in SOEs was around 30 percent for most coastal
provinces, while for other provinces such as Jiangxi, Guangxi, Shaanxi, Gansu, and
Qinghai the shares were still above 70 percent (National Bureau of Statistics, 2009)6.
The above implies that different regions have different skill premia.
Another reason for regional unevenness is different exposure to trade activity.
The policy of opening up was first implemented in the east and led to rapid growth of
exports and FDI inflows. In 1995, the average share of exports in GDP in coastal
provinces was 32 percent, while that of non-coastal provinces was below 7 percent.7
China experienced a sharp increase in trade activity in a relatively short period of time
after its entry into the WTO (Wan et al., 2007; Branstetter and Lardy, 2006), and most
of the increase was in coastal regions.8 By 2007, the average export share in GDP in
6
These shares are calculated using the number of staff and workers in state-owned units divided by the total
number of registered employees. All numbers are for urban areas (National Bureau of Statistics, 2009: table 4-8).
Taking into account those not registered, the SOE shares may overestimate the real shares.
7
Coastal provinces include Beijing, Fujian, Guangdong, Hainan, Hebei, Jiangsu, Liaoning, Shandong, Shanghai,
Tianjin, and Zhejiang. Non-coastal provinces include Anhui, Gansu, Guangxi, Guizhou, Henan, Heilongjiang,
Hubei, Hunan, Jilin, Jiangxi, Inner Mongolia, Ningxia, Qinghai, Shanxi, Shaanxi, Sichuan, Tibet, Xinjiang,
Yunnan, and Chongqing. Export shares are first calculated for each province, and then averaged for coastal and
non-coastal regions. The shares for 2007 are calculated in the same way.
8
Most exports in the export growth regions came from foreign-invested enterprises, Hong Kong-Macau-Taiwan
invested enterprises, and joint ventures. According to our calculation using the Chinese industrial enterprises
database, exports from foreign-invested enterprises accounted for 23 percent of the total exports from firms located
in coastal areas in 2002, and it increased to 35 percent by 2007. Export growth throughout China shows a similar
pattern. Detailed results are available upon request.
7
coastal provinces had reached 45 percent, while that of non-coastal provinces was still
below 8 percent. Along with rapid export growth, China’s exports were becoming
more sophisticated with resources moving from agriculture and textiles into
machinery, electronics, and assembly (Schott, 2006; Amiti and Freund, 2008; Wang
and Wei, 2008). This upgrading of the export sector increased the demand for and
relative wages of skilled labour in coastal regions. While some researchers have
explored the relationship between inequality and China’s globalisation process (Wan
et al. 2007, Zhang and Zhang 2002, Wei and Wu, 2003, Cai et al. 2010), few have
studied the impacts of structural changes brought on by China’s entry into the WTO.
The above regional patterns reflect, in part, the policies of the Chinese
government. In the earlier reform period, the central government offered not only
preferential policies, but also more investment to coastal regions to encourage exports
and attract FDI. The five Special Economic Zones (SEZs)9, for example, are all in
coastal regions (Wang and Wei, 2008). However, in the late 1990s, because of the
widening regional gap, preferential policies were more often designed for the central
and western regions. Many policies were also location-specific, such as policies for
developing the Pudong New District in Shanghai, the Binhai New District in Tianjin,
and the more recent strategy to revitalise the Old Industry Base of the Northeast.
These initiatives often involve preferential policies of the local and central
governments, such as lower income tax rates for foreign invested enterprises, fewer
entry barriers into the finance and service sectors, more favourable treatment for land
9
The five coastal cities are Shenzhen, Zhuhai, Shantou, Xiamen, and Hainan.
8
use permissions, subsidies for fostering industrial upgrading, etc.
Several features of China’s labour market have made this process less effective,
although labour migration still tends to equalise relative wage gaps among regions. In
particular, the Hukou system still imposes high moving costs for migrants. As an
institutional arrangement designed to control population mobility, the Hukou system
requires Chinese citizens to register at a specific location (not necessarily their
birth-place or place of residence) as either the agricultural or non-agricultural type
(Fan, 2008). Hukou status is primarily determined by birth and is passed on from one
generation to the next (Wu and Treiman, 2004). Due to significant differences in
employment opportunities and welfare and benefit entitlements, there is a strong
incentive for rural residents to change their Hukou status from agricultural to
non-agricultural and from countryside to urban areas. However, this change can only
be made through certain channels and complicated procedures, and these channels
generally favour individuals with more skills and/or special achievements.10 For a
migrant with non-agricultural Hukou to change his/her registration place often
involves particular education level requirements and/or employment in industries that
the local government aims to foster. The Hukou quota is often allocated to certain
firm types (e.g., high-tech companies and SOEs). Most coastal employers that attract
large numbers of migrants have little influence in helping employees change their
10
Going to college has been a major channel that increases the probability of a favourable Hukou status. Other
channels that increase this probability include serving in the military, being recruited by SOEs or the government
(Wu and Treiman, 2004; Fan, 2008), rural residents' lands being occupied by urban construction projects (Wong
and Huen, 1998), and rural households purchasing urban housing (Deng and Gustafsson, 2006).
9
Hukou status; as such, only a small number of migrants have successfully changed
their status.
With Hukou reform in recent years, individuals or households have more
freedom and many are choosing to move without changing their Hukou status. In
2005, over 20 percent of the urban workforce was made up of migrants whose Hukou
statuses were not registered in their workplaces, and a vast majority had their Hukou
registered in rural areas.11 Currently, many locally-provided public services, such as
healthcare and schooling, are either unavailable or expensive for these workers. This
raises migration costs for migrants both with agricultural and non-agricultural Hukou.
However, a major shortcoming of our analysis is that rural-to-urban migrants are not
included, as discussed in the data section. We discuss our results' implications for
Hukou reform using the 2005 one percent population survey. The high price of
housing is another factor preventing people from moving from low to high skill
premium locations.
All of the above are also often cited as reasons for growing regional disparity in
income levels, instead of changes in the relative wages of skilled and unskilled labour.
Existing literature has largely neglected the possibility that these regional imbalances
(e.g., in trade policies) and institutional arrangements (e.g., the Hukou system) may
have different impacts on individuals with different skills.12
11
This is according to our calculation using the 2005 one percent population survey.
12
The neglect of regional dispersion of skill premia in previous literature is partly due to limitations of data. Most
existing research emphasises regional disparities based on aggregate data, which makes the discussion of skill
premium difficult. Also, while there is usually a clear relationship between economic development and average
10
3. Data and summary statistics
We use data from three China Household Income Project (CHIP) surveys (1995,
2002, and 2007) to estimate the regional dispersion of skill premia over time. The
CHIP data is known for its high quality and national representativeness. In 1995, 2002,
and 2007, the urban survey covered 11, 12, and 16 provinces or municipalities
respectively, which included a wide variety of regions in terms of geography and
economic development. Individuals in these surveys all have non-agricultural Hukou.
In 1995 and 2002, the surveys collected information from Beijing, Shanxi, Liaoning,
Jiangsu, Anhui, Henan, Hubei, Guangdong, Sichuan, Yunnan, and Guansu—these are
provincial-level divisions, and Chongqing is separated from Sichuan as a municipality
that has been directly controlled by the central government since 1997. One
municipality (Shanghai) and three provinces (Zhejiang, Fujian, and Hunan) joined the
2007 survey. Each province or municipality is treated as a unit, and we use
observations within each unit to estimate provincial skill premium. Municipalities and
cities within provinces are treated as units when we estimate city-level skill premia.
As CHIPs do not follow individuals over time, they do not constitute panel data.
We retain information on males aged 22 to 60 and on females aged 22 to 55.13
We use labour incomes of urban workers, which include regular wages and subsidies
regional income (the regional income level is often used as proxy for its economic development level), the
relationship between relative wages and economic development remains ambiguous.
13
We keep those above 22 years of age because people going to college typically finish at age 22 in China. The
age thresholds of 60 and 55 for males and females reflect current arrangements for retirement age. Our results are
not sensitive to the choice of age thresholds.
11
and other labour income. The terms "wage" and "income" are used interchangeably.
All income data are at the individual level and are deflated using provincial CPIs. The
price levels for all provinces are normalised to one in 1995, and subsequent provincial
CPIs are used to calculate the price levels for 2002 and 2007.14
Because the sampling process of these surveys is based on formal residence
registration (Hukou), the data we use exclude most migrant households in urban areas
without formal residence permits. Considering the large number of rural-to-urban
migrants and the fact that they are usually less educated and worse-off than local
urban workers (Démurger et al., 2009), our data may produce a biased skill premium.
By focusing on a sample with urban Hukou, there is less need to consider the
monetary value of Hukou. Consequently, we are unable to use our findings to infer the
wage compensation differential due to Hukou status.
Table 1 provides summary statistics.15 Both the average age and education level
increased continuously over these years. In 1995, the average age of the urban
working population was 39.5, increasing to 41.5 in 2002, and further to 42.3 in 2007.
The average years of schooling increased from 10.8 in 1995 to 11.3 in 2002 and to
12.7 in 2007. Looking at different education levels a more detailed pattern of
education expansion emerges. The share of the population with tertiary education
increased significantly between 2002 and 2007. Changes in the share of labour force
14
As our focus is the relative income of skilled to unskilled workers in each region, using regional or national
CPIs will not influence our estimate of the skill premia and their distributions.
15
We only report the summary statistics for the observations with positive labour income. Including observations
without income does not change the results significantly, and we do not report them.
12
for different firm ownerships types reflect the restructuring process in urban China. In
the mid-1990s, over 80 percent of the workforce was employed in the public sector in
various forms.16 By 2007, that share had decreased sharply to 53 percent, while the
number of workers in the private sector increased significantly.
Table 2 reports skilled worker shares (columns 1 to 3), income levels (columns 4
to 6), and inequalities (columns 7 to 9) by province. Those with college degrees (3 or
4 years) or above and those with technical school degrees are defined as skilled
workers in this paper, and the others are classified as unskilled workers. As a whole,
the share of skilled workers increased from 40 percent in 1995 to 43 percent in 2002
and to 52 percent in 2007. The pace of this increase differed across provinces; for
example, the share of skilled workers increased by over 20 percent between 2002 and
2007 in Henan, but it barely increased at all in Yunnan.
There were significant increases in income between 1995 and 2007. Annual
average wages increased from 6,209 Yuan in 1995 to 9,772 Yuan in 2002 and further
to 16,173 Yuan in 2007. Data also indicate that both income levels and income growth
were unbalanced across provinces, with coastal regions having higher income levels
and growth rates. In terms of Gini coefficients, there was a significant increase in
16
The public sector includes (1) government employees and (2) workers in SOEs. Between 1995 and 2002, the
shrinking of public sector was mainly due to the reduction of employment in SOEs (from 52 percent in 1995 to 33
percent in 2002). But, we cannot distinguish these two types of employment in 2007. In 1995 and 2002, wage
regressions controlling for a dummy of government or SOE employees produces similar skill premia to those
controlling for two dummies denoting government employees and SOE workers. Therefore, we put these two types
of employment into one category for all three years in this paper.
13
inequality for every province between 1995 and 2002. On the whole, the Gini
coefficient increased significantly from 0.296 to 0.373. For the 2002 to 2007 period,
the Gini of the whole sample increased modestly from 0.373 to 0.407, but the change
differed across regions. For some hinterland provinces, Gini coefficients decreased
(e.g., Shanxi and Henan), while for coastal provinces, Gini coefficients increased (e.g.,
Jiangsu).
4. The regional distribution of skill premia (the returns to education)
4.1 Wage levels of skilled and unskilled workers and skill premia
To see how the income gaps between skilled and unskilled labour (a crude
measure of skill premium) differ across provinces and how the pattern evolves over
time, we report income and income growth for unskilled and skilled workers and their
gaps by province in Table 3. The income gaps were moderate in the mid-1990s,
except for Guangdong. By 2002, the income gap for all provinces had increased
significantly, with some indication that the skilled/unskilled gap in coastal provinces
was larger. Between 2002 and 2007, the gaps in coastal provinces were still growing
faster than those in non-coastal regions. Thus, skilled workers in coastal regions have
higher wages relative to unskilled workers than their non-coastal counterparts.
Regional disparity for unskilled workers also exists, but it is not as noticeable as for
skilled workers. For example, the Beijing-Shanxi income gap for skilled workers
increased sharply from 3,637 Yuan in 1995 to 4,224 Yuan in 2002 and further to 9,942
Yuan in 2007, but the income gap for unskilled workers remained below 3,800 Yuan
14
in all three years. Finally, the results in Table 3 suggest that the increase in income
gaps was mainly due to the faster income growth of skilled workers rather than to the
decline or slow growth in the income of unskilled workers.
The above wage gaps may reflect the effect of other factors, including different
age and gender distributions, ownership structures, and industry compositions.17 We
show the above regularities by estimating the following model:
ln WAGE   0  1 * SKILLED  γX  
(1)
where SKILLED is a dummy variable indicating that an individual has tertiary
education (3 or 4 year college degree or above) or technical school degree. Its
coefficient 1 reflects the extent to which skilled workers earned more than
unskilled.18 Vector X includes experience (= age minus years of schooling minus six),
experience squared, a gender dummy, an ownership dummy, and industry dummies.
We estimate model (1) by province for each year, and the coefficients for
SKILLED are reported in Table 4. For the earlier period (1995 to 2002, columns 1 to
2), the skill premia increased sharply across all regions. This is consistent with
17
Non-work rates may also be different across regions. When samples with wage incomes are not selected from
the working age population at random, the skill premiums using the observed data may be subject to selection bias.
These concerns can be addressed by using Heckman’s two-step selection model (Heckman, 1979), which gives
similar results to OLS; thus, only OLS results are reported.
18
More rigorously, the skilled earn 100*[exp(β1)-1] percent more than the unskilled according to model (1); when
β1 is near zero, exp(β1)-1≈β1. Even when β1 is large and exp(β1)-1>β1, the patterns of the regional distribution of
exp(β1)-1 and β1 will be similar, with exp(β1)-1 showing a wider dispersion. For ease of exposition, we report β1
directly in the text.
15
existing literature that documents an overall increasing trend in the return to education
(Liu et al., 2010; Zhang et al., 2005). Less well documented is the large variation
across regions in skill premia. In 1995, the skill premium in Guangdong was the
highest (0.399), while that of Beijing was the lowest (0.132). The second lowest was
Yunnan (0.153) followed by Sichuan (0.165). In 2002, Beijing remained the province
with the lowest skill premium (0.249) and Gansu became the province with the
highest skill premium (0.473), followed by Liaoning (0.470), Guangdong (0.430), and
Sichuan (0.430). Gansu and Sichuan are western provinces, and for these there seems
to be no clear relationship between skill premia and economic development. For the
period between 2002 and 2007 (columns 2 to 3), there are signs that skill premia
stopped rising or even declined in central and western provinces (see Shanxi,
Chongqing, and Gansu for example). For coastal regions skill premia kept rising.
4.2 The return to years of schooling and the returns to different education levels
Dividing the labour force into skilled and unskilled categories is a simplification,
since along with educational expansion in the late 1990s there were significant
composition changes within each group. In the unskilled group the relative number of
high school graduates increased significantly, as did the relative number of college
graduates in the skilled group (see Table 1). Estimated skill premia may thus be
contaminated by these changes. Therefore, we estimate the returns to years of
schooling and the returns to various levels of education using two different models:
ln WAGE   0  1 * SCHOOLING  γX  
(2)
16
ln WAGE   0    i * EDU i  γX  
(3)
i
In model (2) SCHOOLING is years of formal schooling, and in model (3) EDUi
are dummies for various education levels. X controls for the same factors as in model
(1). Results for these models are reported in the remaining columns of Table 4, and
generally, the patterns of these results are consistent with our results as presented
above. For the returns to years of schooling (columns 4 to 6), the most important
feature is the large regional variation in 2007. While the returns are as high as 10 to
14 percent in coastal regions, they are well below 10 percent in non-coastal provinces.
The standard deviation of these provincial returns was 0.024 in 2007, doubling that of
2002 and more than tripling that of 1995 (see last row of Table 4).
The college premia show similar patterns, but their dispersion increased only
slightly (columns 7 to 9). For space reasons the returns to other education levels are
not reported, though some features of these returns are worth noting. First, most of the
increases in the returns to education are concentrated in higher levels of education,
especially college education. Second, coastal regions witnessed rising returns for
almost every tertiary education level. Finally, the returns to high school decreased
between 1995 and 2007.
4.3 Skill premia at the city level
We estimate wage equations for 62, 61, and 91 cities for 1995, 2002, and 2007
respectively, using the same specification as in model (1). After calculating the skill
premia, we estimate their yearly kernel densities non-parametrically, each city being
17
treated as an observation without using city population or employment as weight. We
use the Epanechnikov kernel function and the bandwidth set at 0.08. The distributions
are reported in Figure 1.19 Apparently they are non-degenerate in all three years. The
distribution moved to the right and became more dispersed between 1995 and 2002.
In the second period (2002 to 2007) it moved slightly to the right. A fatter right tail is
observed in the 2007 distribution, which is likely explained by the fairly high skill
premia of some coastal cities. The regional distribution for the return to years of
schooling evolves in a similar way, which we do not report due to space limitations.
In the following section, we assess the importance of various factors behind the
evolution of these city skill premia and returns to schooling.
5. Explanation of results and their implications
5.1 Factors that influence the evolution of the regional skill premium
In this section, we estimate the following fixed effects model:
S K I L L P Ri Et M0  1*
E X Pi t G
 D2*P
SO
3* E
i t
SγX
+K
i(4)
i t I Li L
t E D
i t
In this model, subscript i refers to city, and each city is an observation in the
estimation; t=1995, 2002, or 2007 refer to the survey year. SKILLPREM is the
estimated skill premium or return to years of schooling in section 4.3, and (unlike
wages in wage equations) it is not specified in log form. We use the ratio of exports to
19
We also estimate kernel densities for the distribution of city skill premia for cities common to all three surveys.
While only 33 cities remained and the estimates became less accurate, the regional distribution evolved in a similar
pattern. We do not report it to save space.
18
u
GDP, EXPGDPit, to measure the openness of city i at time t. SOEit is the share of the
labour force in state-owned enterprises or governments. SKILLEDit is the share of
skilled labour in the labour force. Vector X controls for wage levels and age structures.
ui is a time invariant unobservable regional characteristic. To keep the panel balanced,
we use the 33 cities that participated in all three surveys in the regressions.
The results are reported in Table 5. In columns 1 to 3 we use the returns to years
of schooling as the dependent variable. Column 1 shows the results when all three
years are used. The only coefficient that is significant is for SOE, suggesting that
ownership restructuring played a significant role between 1995 and 2007; however,
some factors may have different impacts over time. Columns 2 and 3 report the results
for the two periods of 1995 to 2002 and 2002 to 2007. In the first period, export
activity did not play a significant role. The coefficient on EXPGDP is small (-0.021)
and its standard error is relatively large (0.047). For the 2002 to 2007 period, the
coefficients on openness (EXPGDP) become significantly positive. A 10 percent
increase in export share was associated with a 1.5 percent increase in the return to
schooling. The change in the coefficient on the export variable indicates a structural
change after China’s entry into the WTO. Indeed, export volumes increased sharply in
this period, trade activity became more regionally concentrated, and export products
became more sophisticated. These factors increased the demand for skilled labour in
coastal regions and caused a wider regional dispersion of the returns to education.
Equally interesting are the coefficients for SOE. Between 1995 and 2002, the
coefficient of SOE was significantly negative, which means that the relative reduction
19
in the size of the public sector played a major role in the increase of the returns to
education. This result is consistent with the fact that the private sector had higher
returns than the public sector. The coefficients of SOE only became significant (at the
10 percent level) between 2002 and 2007, although the absolute value increased.
Considering that the SOE shares decreased more sharply in the first period (see Table
1), ownership restructuring played a larger role in the former than in the latter period.
Columns 4 to 6 use city skill premia as the dependent variable and lead to similar
conclusions.
Due to data limitations, we cannot explore all of the factors that may affect
regional skill premia. One unexplored hypothesis is that high migration rates of
unskilled labour slow the growth of real wages in coastal regions and drive up skill
premia there. We use information from the 2005 one percent population survey to
shed light on this. Figure 2 plots the relationship between skill premia and the migrant
shares of skilled and unskilled workers in the urban labour market. We run a
regression of log of wage on a skill dummy for each province to calculate skill
premium, controlling for age, age squared, and gender. In this dataset, the skilled
workers are those with degrees higher than high school, and the migrants are those
who have left their Hukou registration location to work in another city more than half
a year. Figure 2 shows that the migrant shares of unskilled workers are higher than the
migrant shares of skilled workers, especially in coastal provinces. However, there is
only a weak positive correlation between the migrant share of unskilled workers and
skill premium (bottom panel of Figure 2), indicating that the inflow of unskilled
20
workers is not a significant factor in increases of coastal skill premia.20 On the
contrary, a more significant and larger positive correlation between the migrant share
in skilled workers and skill premium is observed in the top panel of Figure 2,
supporting our demand-driven conclusion. 21 As this is often policy related, we
discuss the relative inelastic supply of skilled workers in the policy implications
section (section 5.3).
5.2 Skill premia and inequality
One implication of the evolution of regional distributions of skill premia is in
relation to regional inequality. In Figure 3, we plot the relationship between skill
premia and wage inequality at the provincial level. In 1995, both skill premia and
wage inequalities at the provincial level were low. Between 1995 and 2002, there was
a general increase in skill premia across regions. As a consequence, within-region
inequalities increased dramatically. Between 2002 and 2007, the dispersion of skill
premia across regions also grew, as did inequalities across different regions. In 2007,
the relationship between skill premia and inequalities became more positive and
significant. We also run regressions of regional Gini coefficients on skill premia in
different years; in all of these regressions, the coefficients on skill premia are positive.
20
Regressing provincial skill premia on migrant shares of the unskilled urban workers results in a coefficient of
0.110(0.119). Dropping Yunnan province produces a coefficient of 0.290(0.173). The adjusted R-squared are
-0.0052 and 0.0607 respectively.
21
Regressing provincial skill premia on migrant shares of skilled urban workers, we get a coefficient of
0.228(0.066). Dropping Yunnan province produces a coefficient of 0.381(0.090). The adjusted R-squared are
0.2730 and 0.3740 respectively. Tibet is dropped in all regressions.
21
In the years 1995 and 2002, the coefficients are relatively small and insignificant;
however, in 2007 the coefficient is higher and significant at a 1 percent significance
level.22
Given the positive correlation between regional skill premium and within-region
inequality, we can assess the implications of the evolution of the regional skill
premium for inequality. First, as skill premia increased significantly across regions, so
did within-region inequalities, and it played a progressively important role in
determining the level of overall urban inequality and its change. 23 Second, the
distribution pattern of the regional skill premia indicates that coastal regions have
much higher within-region inequalities. Our exercise shows that within-region
inequality in coastal regions has increased more than in non-coastal regions, and
contributes more to the increase of overall urban wage inequality.24
22
These coefficients are 0.107(0.086), 0.100(0.082, and 0.221(0.047), in 1995, 2002, and 2007, respectively.
23
Formally, we decompose the overall urban wage inequality into between- and within-province inequality. To
guarantee that the inequality is regionally decomposable, we use an inequality measure of GE(2) (Generalised
Entropy Index with sensitivity parameter 2). This decomposition can be found in Gustafsson et al. (2008). In 1995,
the overall and within-region inequalities measured as GE(2) were 0.181 and 0.140. Between 1995 and 2002,
almost all of the increase in overall wage inequality (from 0.181 to 0.284) came from within-region inequality
(from 0.140 to 0.243) and between-region inequality decreased slightly. Again, most of the inequality change
between 2002 and 2007 (from 0.284 to 0.375) is attributable to within-region inequality (from 0.243 to 0.321).
Dropping the four provinces added in 2007 did not change this pattern.
24
Between 2002 and 2007, the decomposition exercises for the coastal and non-coastal regions show that the
overall GE(2) increased by 0.034 (=0.227-0.193) in non-coastal regions, and that of coastal regions increased by
0.055 (=0.361-0.306). These differential changes are mainly due to the differential changes in within-province
inequality, which are 0.032 (=0.223-0.191) and 0.063 (=0.332-0.269) in non-coastal and coastal regions,
respectively. Again, dropping the four provinces added in 2007 did not change this pattern.
22
5.3 Policy implications
Our results also have implications for many policy areas, including the Hukou
system, housing prices, training and retraining programmes, and the unemployment of
college graduates. Skill premia in coastal regions were relatively higher than those in
non-coastal regions, and this differential was more significant in 2007 than in earlier
years, which indicates a stronger demand for skilled labour. That the skill premia have
not been equalised across provinces indicates that across-province mobility of skilled
labour is incomplete. Figure 2 and Figure 4 provide further evidence supporting this
conclusion. The upper panel of Figure 2 shows that around one-third of the skilled
workers are migrants in Beijing, Shanghai, and Guangdong, and over 10 percent of
skilled workers in 23 provinces are migrants; these shares are much lower than those
for unskilled workers, especially in coastal regions (see the bottom panel of Figure 2).
Taking into consideration the large difference in the number of skilled and unskilled
workers, skilled migrants are much lower in number than unskilled migrants.25 Figure
4 reports the relationship between these shares and real wages of skilled and unskilled
workers. The results suggest that skilled workers are less responsive to regional
income differentials than unskilled workers. While wage differences are much larger
for skilled workers, their migrant share differences are lower than the share for
25
According to our calculation based on the 2005 one percent population survey, the ratio of unskilled to skilled
migrants is 5.8 to 1 in the urban labour market. The migrants include workers with both agricultural and
non-agricultural Hukou. It is worth noting that there is a larger share of migrants among skilled workers than
among unskilled workers. The migration rates for unskilled and skilled workers are 7.5 percent and 15 percent,
respectively. Detailed results are available upon request.
23
unskilled workers.26
These results point to how college graduates should be allocated under the
educational expansion policy, favouring coastal regions over non-coastal regions.
Factors that increase the moving cost include the Hukou system and high housing
prices. Although working in a city where they are not registered as permanent
residents is not officially restricted, obtaining permanent local residence status
remains difficult. As most new college graduates in a city do not have housing, they
either pay high rents or live in poor conditions, such as dormitories in suburban areas.
Figure 5 shows a positive correlation between skill premia and housing prices,
suggesting that high housing prices are a factor that retards mobility, especially for
skilled workers. 27 To weaken these mobility-retarding factors, the household
registration system needs to change, as does the housing supply system. Emphasising
the negative effects of larger cities neglects the high returns to college graduates.
Could this type of regional development widen skill differentials rather than
shrink them? This seems unlikely, as market forces tend to equalise skill premia
across regions if migration costs are sufficiently low. As shown in Table 3 for 2007,
skilled workers had larger regional wage differentials than unskilled workers and
26
Regressing provincial incomes of the skilled workers on migrant shares of skilled urban workers we get a
coefficient of 0.00017(0.000015), much smaller than the results for unskilled workers, 0.00068(0.00008).
Dropping Yunnan province produces 0.00017(0.000014) and 0.00072(0.000080), respectively. Tibet is dropped in
all regressions.
27
It is possible that high housing prices are the result of more skilled workers moving to large cities and other
factors as well, but identifying the direction of causality is beyond the scope of this paper.
24
more incentive to move to high-premium regions. Also, as shown in Figure 2, the
relative number of skilled migrant workers is still low in the urban labour market.
Therefore, skilled workers have greater potential to move than unskilled. The results
in Table 6 indicate that skilled workers have a stronger preference for permanent
housing, social security, and public services. Using the 2005 one percent population
survey, Column 1 shows that skilled workers are more likely to purchase permanent
housing rather than to rent an apartment or live at construction site dormitories,
though this is influenced by personal and family characteristics including age,
monthly income, gender, marital status, ethnicity, family size, and city dummies.
Using the floating population monitoring data for 2012 compiled by the National
Population and Family Planning Commission, columns 2 and 3 show that skilled
migrants are more willing to change their Hukou registration to their workplaces and
to settle down. On the basis of this evidence, we conjecture that reforming the Hukou
system and increasing housing supply will encourage more skilled workers to move.
Regional development would lead to a continuing demand for infrastructure
investment as larger populations become concentrated in high-income cities. At some
stage such investment will face constraints determined by the endowments of a
specific area and suffer diminishing returns. These constraints and negative
externalities also mean that increasing concentrations of workers in high-wage regions
could cause problems related to traffic congestion, pollution, and environmental
pressures in labour receiving cities. If skilled workers demand higher compensation
for these disamenities, skill premia could remain high in these regions. However, in
25
this case, high skill premium is associated with a smaller inequality in utility.
Moreover, as the labour market becomes more flexible labour is allocated more
efficiently, which is beneficial for high growth. Given the fact that Chinese cities are
undersized (Au and Henderson, 2006), encouraging skilled workers to move by
reforming Hukou and increasing housing supply will induce more returns than costs.
Meanwhile, the local governments should improve their management capacity to
reduce congestion and pollution costs.
The high premia in some regions also suggests a role for the local government
and employers, including the development of education and training programmes.
Take Yunnan as an example: Its skill premium increased from 0.307 to 0.483 between
2002 and 2007, but its share of skilled workers remained almost unchanged. In this
case, increasing the education level and providing a training programme for unskilled
workers may reduce the skill premium and inequality. Investing in and offering
preferential policies to low premia regions may also help equalise skill premia across
regions; however, location- or industry-specific policies often involve more
complicated political processes, and whether this can reduce the dispersion in skill
premia and inequality depends on local circumstances, technology, the nature of the
policies, and labour mobility.
6. Conclusions
This paper uses three Chinese household surveys (1995, 2002, and 2007) to
estimate the levels and changes in skill premia for different provinces and cities. We
26
use the surveys to estimate regional dispersion in skill premia. Although there is an
increase in skill premia in nearly all provinces between 1995 and 2002, we do not find
such increases between 2002 and 2007. Data from the 2007 survey indicate a strong
positive relationship between openness and skill premia, with coastal regions having
much higher skill premia than non-coastal regions. We interpret the 2002 to 2007
results as being reflective of China’s deepening globalisation after entry into the
WTO.
These results also have policy implications. As the employment of skilled
workers (college graduates in particular) is an extremely important force for China’s
urbanisation process, reducing moving and settling costs is needed to sustain high
levels of growth and to mitigate inequality. Labour-losing areas may face the
phenomenon of brain drain caused by the out-migration of skilled workers. The
central government can play a role in balancing economic development by making
fiscal transfers from the well-developed to the less-developed regions, and by
providing quality public service in education.
In using the CHIPs to estimate the skill premia we exclude most migrants in the
labour market. Investigating the compensation differentials regarding Hukou status
(i.e., the migrant's marginal willingness to pay for the welfare and benefit entitlements
associated with the Hukou status of his/her workplace) is beyond the scope of this
paper and we leave it for future research. However, we conjecture that compensation
differentials regarding Hukou status is unlikely to explain the increase in the
dispersion of regional skill premia, as there is no concrete evidence that Hukou value
27
increases in high premium regions.
References
Amiti, Mary, and Freund Caroline. 2008. “An Anatomy of China’s Export Growth,” NBER
Working Paper.
Appleton, Simon, John Knight, Lina Song, and Qingjie Xia. 2002. “Labor Retrenchment in China:
Determinants and Consequences,” China Economic Review, 13(2-3): 252-275.
Au, Chun-Chung, and J. Vernon Henderson. 2006. “Are Chinese Cities Too Small? ” Review
of Economic Studies,73: 549-576.
Branstetter, Lee and Nicholas Lardy. 2006. “China's Embrace of Globalization,” NBER Working
Paper No. 12373.
Cai, Fang, Yang Du, and Meiyang Wang. 2009. “Migration and Labor Mobility in China,” Human
Development Research Paper, 2009/09.
Cai, Hongbin, Yuyu Chen, and Li-an Zhou. 2010. “Income and Consumption Inequality in Urban
China: 1992-2003,” Economic Development and Cultural Change, 58: 385-413.
Chen, Jian, and Belton M. Fleisher. 1996. “Regional Income Inequality and Economic
Growth in China,” Journal of Comparative Economics, 22(2): 141–164
Deng, Q, and Gustafsson B. 2006. “China's Lesser Known Migrants.” IZA Discussion Papers
2152, IZA.
Démurger, S, Gurgand M, Li S, and Yue X. 2009. “Migrants as Second-Class Workers in Urban
China? a Decomposition Analysis.” Journal of Comparative Economics 37(4): 610-628
Dong, Xiaoyuan and P. Bowles. 2002. “Segmentation and Discrimination in China’s Emerging
Industrial Labor Market,” China Economic Review, 13(2-3): 170-196.
Fan, C. 2008. China on the Move: Migration, the State, and the Household. Routledge, London
and New York.
Gustafsson, B, Shi Li, Terry Sicular, and Ximing Yue. 2008. “Income Inequality and Spatial
Differences in China, 1988, 1995, and 2002,” in Gustafsson B, Shi Li, Terry Sicular (ed),
Inequality and Public Policy in China, Cambridge University Press.
Heckman, J. J. 1979. “Sample Selection Bias as a Specification Error,” Econometrica, 47.
pp.153-161
Hsieh, Chang-Tai; Klenow Peter J. 2009. "Misallocation and Manufacturing TFP in China and
India," Quarterly Journal of Economics 124, November 2009, 1403-1448.
Li, Shi and Sai Ding. 2003. “Long-term Change in Private Returns to Education in Urban China,”
Social Science in China, No. 6. (Chinese)
Li, Yao, John Whalley, Shunming Zhang, and Xiliang Zhao. 2011. “The Higher Educational
Transformation of China and its Global Implications,” World Economy, 34(4): 516-545.
Liu, Xuejun, Albert Park, and Yaohui Zhao. 2010. “Explaining Rising Returns to Education in
Urban China in the 1990s,” IZA Discussion Paper 4872.
Meng, Xin. 2000. Labor Market Reform in China, Cambridge University Press
National Bureau of Statistics. 2008/2009. China Statistical Yearbook. China Statistics Press.
Schott, Peter. 2006. “The Relative Sophistication of Chinese Exports,” NBER Working Paper No.
12173.
28
Solinger, D. J. 2002. “Labor in Limbo: Pushed by the Plan toward the Mirage of the Market,” In F.
Mengin and Jean-Louis Rocca (eds.) Politics in China, Palgrave.
Wan, Guanghua, Ming Lu, and Zhao Chen. 2007. “Globalization and Regional Income Inequality
in China, Empirical Evidence from within China,” Review of Income and Wealth, 53(3):
35-59
Wang, Zhi, and Shang-Jin Wei. 2008. “What Accounts for the Rising Sophistication of China’s
Exports?” NBER Working Paper No. 13771.
Wei, Shang-Jin and Yi Wu. 2003. “Globalization and Inequality: Evidence from within China,”
CEPR Discussion Paper No. 3088.
World Bank. 2009. “From Poor Areas to Poor People: China's Evolving Poverty Reduction
Agenda”, Report No. 47349-CN.
Wong, L, and Huen W. 1998. ‘Reforming the Household Registration System: a Preliminary
Glimpse of the Blue Chop Household Registration System in Shanghai and Shenzhen.”
International Migration Review 32(4): 974-994
Wu, X, and Treiman D. 2004. “The Household Registration System and Social Stratification of
China, 1955-1996.” Demography 41(2):363-384
Xing, Chunbing. 2008. “Human Capital and Wage Determination in Different Ownerships,
1989-97,” in Wan Guanghua (ed.), Understanding Inequality and Poverty in China: Methods
and Applications, Palgrave Macmillan, 117-135.
Zhang, Xiaobo, and Kevin Zhang. 2003. “How Does Globalization Affect Regional Inequality
within a Developing Country? Evidence from China,” Journal of Development Studies 39(4):
47-67.
Zhang, Junsen, Yaohui Zhao, Albert Park, and Xiaoqing Song. 2005. “Economic Returns to
Schooling in Urban China, 1988 to 2001,” Journal of Comparative Economics, 33: 730-752.
29
Table 1. Summary statistics
1995
2002
2007
(1)
(2)
(3)
Age
39.53
41.51
42.30
Female
0.468
0.457
0.457
Years of schooling
10.78
11.27
12.72
Middle school and below
0.358
0.287
0.221
High school
0.240
0.279
0.262
Technical school
0.161
0.119
0.113
3-year college
0.160
0.217
0.253
4-year college and above
0.082
0.098
0.150
SOE and government
0.814
0.612
0.532
No. of obs.
11189
10774
15249
Variable
Note: The SOE sector includes state enterprises or joint-owned enterprises with the State being the largest share
holder at the central, provincial, or city level.
30
Table 2. Income and income inequality in urban China by province, 1995-2007
Skilled Worker Share
Average Annual Income
1995 Yuan
Gini Coef.
1995
2002
2007
1995
2002
2007
1995
2002
2007
Province
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
(9)
Beijing*
51.6
47.7
61.2
8457
12044
20348
0.243
0.348
0.365
Shanxi
43.4
44.0
49.9
4822
8132
12405
0.276
0.363
0.325
Liaoning*
39.2
41.0
48.6
5522
9216
11641
0.267
0.353
0.383
Shanghai*
Jiangsu*
46.8
32.9
36.4
Zhejiang*
Anhui
45.4
24490
6742
10350
43.7
34.0
41.8
Fujian*
47
18986
0.422
0.262
0.367
21990
4900
8250
51.9
12779
0.446
0.401
0.253
0.346
15792
0.338
0.372
Henan
38.8
42.2
62.6
4737
7746
11973
0.278
0.326
0.313
Hubei
41.3
51.3
55.3
5866
8532
14140
0.246
0.326
0.358
Hunan
53
12830
0.351
Guangdong*
36.3
41.2
52.7
11201
17929
26246
0.307
0.376
0.403
Chongqing
42.8
40.3
48.4
4762
9351
12473
0.232
0.340
0.348
Sichuan
41.6
36.9
51.4
5940
8068
12248
0.258
0.361
0.381
Yunnan
45.6
51.7
52.3
5733
9186
11821
0.202
0.301
0.366
Gansu
37.4
43.1
48.7
4604
7641
9807
0.241
0.344
0.371
Total
40.2
43.4
51.6
6209
9772
16173
0.296
0.373
0.413
(52.4)
(15522)
(0.412)
Note: Skilled labour includes graduates of technical school, professional college (3 year college), and college (4
year) or above. Unskilled labour includes high school graduates or those with degrees below these levels. Columns
1-3 calculate the share of skilled labour only for those with wage incomes. Wages are deflated using provincial
level urban CPIs. The price levels for all provinces are normalized to one in 1995. Numbers in parenthesis are
calculated excluding observations from the newly added provinces. Coastal provinces are marked with an asterisk
"*".
31
Table 3. Income and income growth for unskilled and skilled workers in urban China by province, 1995-2007
Unskilled workers
Skilled workers
Average Annual Income
Changes
1995
2002
2007
95-02
Province
(1)
(2)
(3)
Beijing*
7939
10038
Shanxi
4452
Liaoning*
4982
Average Annual Income
Changes
02-07
1995
2002
2007
95-02
02-07
1995
2002
2007
(4)
(5)
(6)
(7)
(8)
(9)
(10)
(11)
(12)
(13)
14313
2099
4275
8942
14244
24170
5302
9926
1003
4206
9857
6648
10588
2197
3939
5305
10020
14228
4716
4207
853
3372
3640
7510
9095
2528
1585
6360
11673
14335
5313
2662
1378
4163
5240
Shanghai*
Jiangsu*
16551
6124
8086
Zhejiang*
Anhui
12300
6711
Fujian*
10669
Skilled - Unskilled
33508
1962
4213
8003
14305
15351
4469
Income Gap
27035
16957
6302
12730
1879
6219
30531
2242
3958
5736
10394
11831
15164
14735
15180
4658
4770
1267
3683
19470
4495
7639
Henan
4214
6352
9391
2139
3039
5564
9657
13516
4093
3859
1350
3305
4125
Hubei
5346
6902
11574
1556
4672
6606
10077
16218
3471
6141
1260
3175
4644
Hunan
9445
15836
6391
Guangdong*
9700
14583
19141
4883
4558
13841
22707
32613
8866
9906
4141
8124
13472
Chongqing
4114
7129
10126
3014
2998
5630
12642
14971
7012
2329
1516
5513
4845
Sichuan
5391
6314
8579
923
2265
6713
11073
15713
4360
4639
1322
4759
7134
Yunnan
5326
7044
8227
1718
1184
6218
11184
15101
4966
3917
892
4140
6874
Gansu
4230
5868
7276
1637
1409
5230
9978
12469
4748
2491
1000
4110
5193
Total
5621
7859
11903
2238
4044
7083
12270
20172
5187
7902
1462
4411
8269
(11479)
(3620)
(19192)
(6922)
(7713)
Note: Skilled labour includes graduates of technical school, professional college (3 year college), and college (4 year) or above. Unskilled labour includes high school graduates or those with
degrees below these levels. Wages are deflated using provincial level urban CPIs. The price levels for all provinces are normalized to one in 1995. Numbers in parenthesis are calculated
excluding observations from the newly added provinces. Coastal provinces are marked with an asterisk "*".
32
Table 4. Provincial skill premia, return to education, and college premium in urban China, 1995-2007
Skill Premium
Return to Years of Schooling
College Graduate Premium
1995
2002
2007
1995
2002
2007
1995
2002
2007
Province
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
(9)
Beijing*
0.132
0.249
0.485
0.026
0.059
0.114
0.207
0.414
0.618
Shanxi
0.184
0.340
0.289
0.037
0.059
0.066
0.193
0.315
0.470
Liaoning*
0.319
0.470
0.350
0.048
0.075
0.078
0.333
0.602
0.510
Shanghai*
Jiangsu*
0.683
0.170
0.340
Zhejiang*
Anhui
0.597
0.135
0.034
0.058
0.606
0.226
0.306
Fujian*
0.244
0.117
0.895
0.057
0.537
0.114
0.031
0.072
0.351
0.053
0.657
0.574
0.248
0.401
0.091
0.345
0.442
Henan
0.218
0.263
0.363
0.039
0.056
0.066
0.325
0.445
0.504
Hubei
0.176
0.314
0.326
0.027
0.046
0.080
0.334
0.365
0.499
Hunan
0.340
0.077
0.513
Guangdong*
0.399
0.430
0.466
0.037
0.084
0.117
0.433
0.684
0.700
Chongqing
0.244
0.398
0.354
0.039
0.056
0.066
0.164
0.559
0.587
Sichuan
0.165
0.430
0.440
0.028
0.058
0.089
0.199
0.495
0.640
Yunnan
0.153
0.315
0.480
0.027
0.047
0.090
0.219
0.332
0.566
Gansu
0.277
0.473
0.408
0.039
0.074
0.072
0.367
0.504
0.457
Standard
0.078
0.077
0.007
0.012
0.024
0.104
0.113
0.127
Deviation
0.123
(0.098)
(0.022)
(0.101)
Note: For each province, we run an OLS regression to get the skill premium and the return to education. We also control for experience, experience squared, gender, industry and ownership
dummies. In the first three columns, skilled labour includes those with college degrees (3 or 4 years) or above and technical school graduates. Those with high school degrees or below are
classified as unskilled workers. Numbers in parenthesis are calculated excluding the newly added provinces. Coastal provinces are marked with an asterisk "*".
33
Table 5. Explaining the regional variation of the returns to education and skill premia
Dependent var: return to yrs of schooling
Dependent var: skill premium
1995-2007
1995-2002
2002-2007
1995-2007
1995-2002
2002-2007
(1)
(2)
(3)
(5)
(5)
(6)
0.030
-0.021
0.152**
0.200
0.072
0.917**
(0.051)
(0.047)
(0.065)
(0.211)
(0.176)
(0.400)
-0.170***
-0.134**
-0.172*
-0.643***
-0.340
-0.796
(0.045)
(0.064)
(0.096)
(0.233)
(0.286)
(0.586)
-0.009
0.007
-0.002
-0.030
0.020
0.052
(0.028)
(0.032)
(0.063)
(0.147)
(0.160)
(0.292)
0.048
0.051
-0.055
0.185
0.185
0.463
(0.064)
(0.050)
(0.163)
(0.301)
(0.268)
(0.606)
0.087
0.080
0.042
-0.037
-0.132
-1.763**
(0.057)
(0.075)
(0.164)
(0.342)
(0.373)
(0.801)
0.044
0.059
0.166
-0.219
-0.515
-0.752
(0.051)
(0.073)
(0.168)
(0.282)
(0.359)
(0.736)
-0.001
0.020
-0.037
-0.508
-0.754
-1.244
(0.070)
(0.117)
(0.183)
(0.430)
(0.451)
(1.041)
0.179
0.012
0.118
1.022
0.557
0.745
(0.251)
(0.301)
(0.521)
(1.319)
(1.512)
(2.381)
R2_wthin
0.420
0.508
0.289
0.353
0.611
0.187
N
99
66
66
99
66
66
EXPGDP
SOE
SKILLED
EXPER11-20
EXPER21-30
EXPER31+
MWAGE
Constant
Note: EXPGDP is the ratio of exports to GDP; SOE is the share of the labour force in state owned enterprises or governments.
SKILLED is the share of skilled labour in the labour force of a city. EXPER11-20 are the shares of the working population
with potential experience ranging from 11 to 20, EXPER21-30 and EXPER31+ are defined similarly. MWAGE is the average
wage level. We run the above regressions using a fixed effects model. *, **, and *** are significance levels at 10%, 5%, and
1% respectively. Standard errors are in parenthesis.
Sources: China Statistic Yearbook and authors calculations from CHIPs.
Table 6. Migrants' propensity to purchase housing and their willingness to settle down and change Hukou
2005 1% population survey
2012 floating population monitoring data
purchase permanent housing
willing to stay over 5 years
willing to change Hukou
(yes=1/no=0)
(yes=1/no=0)
location (yes=1/no=0)
(1)
(2)
(3)
0.122***
0.056***
0.060***
(0.006)
(0.004)
(0.005)
Pseudo R2
0.274
0.080
0.100
N
76934
132623
132684
skilled
Note: All regressions are run for migrant household head, controlling for cubic age, gender, ethnicity, Hukou status, marital
status, quadratic log of monthly income, and city dummies. Column 1 also controls for family size, but columns 2 and 3 do
not for the lack of data.
The skilled are individuals with college degrees (3 or 4 years) or above and technical school
graduates. Probit models are estimated and the marginal effects are reported.
34
3
Figure 1. Distribution of skill premia at city level
0
1
Density
2
1995
2002
2007
-.4
0
.4
.8
Skill Premium
1.2
1.6
2
Note: The skill premia are obtained by estimating wage equation for each city. There are 62, 61, and 91 cities for 1995, 2002,
and 2007. The kernel density for each year is estimated using the Epanechnikov kernel function, bandwidth is set at 0.08, and
each city is treated equally without using city population or employment as weight.
35
.4
.5
Figure 2. The relationship between skill premia and migrant shares of skilled and unskilled workers
Shanghai
Beijing
.3
Guangdong
Fujian
.2
Zhejiang
Jiangsu
Qinghai
Yunnan
Hainan
Sichuan
Liaoning
Anhui
Hunan
ShandongChongqing
Guizhou
HubeiGuangxi
Shaanxi
Jiangxi
Henan
Hebei
Ningxia
Xinjiang
InnerM Jilin
Gansu
Shanxi
HLJ
0
.1
Tianjin
.4
.8
1.2
1.6
Skill Premium
.5
Shanghai
Beijing
Guangdong
.4
Fujian
.3
Zhejiang
Tianjin
Xinjiang
.2
InnerM
Qinghai
Jiangsu
Guizhou
Hainan
Guangxi
Yunnan
0
.1
Liaoning
Hunan
Hubei
Ningxia
Sichuan
Shaanxi Chongqing
Anhui
Jilin
HLJ Hebei
Gansu
Shanxi
Shandong
Henan
Jiangxi
.4
.8
1.2
1.6
Skill Premium
Note: We regress log of wage on a skill dummy to get the skill premium for each province, controlling for age, age squared,
and gender. Wages are deflated using provincial level urban CPIs. The price levels for all provinces are normalized to one in
1995. Skilled workers are those with a degree higher than high school and with positive wage income. Migrants are those
who have left their Hukou registration location and work in another city more than half a year.
Sources: One percent population survey for 2005, China Statistic Yearbook 1997-2006.
36
.45
Figure 3. The relationship between skill premia and within province inequality
.35
.25
.3
2002
1995
.2
Gini Coef.
.4
2007
0
.2
.4
Skill Premium
.6
.8
Note: The skill premium for each province is estimated according to model (1) using data from CHIP 1995, 2002, and 2007.
Regressing Gini coefficients at the provincial level against skill premium, the coefficients on skill premium are 0.107(0.086),
0.100(0.082), and 0.221(0.047) for 1995, 2002, and 2007, respectively. The adjusted R-squared are 0.046, 0.041, and 0.582,
respectively. The numbers of observations are 12, 12, and 16.
37
.4
.5
Figure 4. The relationship between real wages and migrant shares of skilled and unskilled workers
.3
Beijing
Shanghai
Guangdong
Fujian
.2
Yunnan
Zhejiang
0
.1
HainanJiangsu
Qinghai
Sichuan
Chongqing
Liaoning
Anhui
Tianjin
Hunan
Shandong
Guizhou
Guangxi
Hubei
Shaanxi
Jiangxi
Henan
Hebei
Ningxia
Xinjiang
InnerM
Jilin
Gansu
Shanxi
HLJ
500
1000
1500
2000
Average Wage for Skilled Workers
2500
.5
Shanghai
.4
Beijing
Guangdong
Fujian
.3
Zhejiang
InnerM
Guizhou
Hainan
Guangxi
Yunnan
Liaoning
Hunan
Hubei
Ningxia
Sichuan
Shaanxi
Anhui
Chongqing
Jilin
HLJ
Hebei
Gansu
Shandong
Shanxi
HenanJiangxi
0
.1
.2
Qinghai
Tianjin
Xinjiang
Jiangsu
500
1000
1500
2000
Average Wage for Unskilled Workers
2500
Note: Wages are deflated using provincial level urban CPIs. The price levels for all provinces are normalized to one in 1995.
Skilled workers are those having degrees higher than high school graduates and with positive wage income. Migrants are
those who have left their Hukou registration location and work in another city more than half a year.
Sources: One percent population survey for 2005, China Statistic Yearbook 1997-2006.
38
Figure 5. The relationship between skill premia and housing prices
8000
10000
Beijing
6000
Shanghai
Guangdong
Zhejiang
4000
Fujian
Jiangsu
2000
Anhui
.2
Liaoning
Hubei
Sichuan
Chongqing
Henan Gansu
ShanxiHunan
.3
Yunnan
.4
.5
Skill Premium
.6
.7
Note: The skill premium for each province is estimated by running a wage equation according to model (1) using CHIP2007.
The housing prices are provincial averages for commercial residence housing.
Sources: China Statistic Yearbook 2008 and authors calculations.
39